Estimation of the hydraulic conductivity of bentonite–polymer GCLs with machine learning techniques

Machine learning models were developed to predict and classify the hydraulic conductivity of bentonite–polymer geosynthetic clay liners (B-P GCLs). The dataset comprises 316 hydraulic conductivity tests from the literature, along with associated factors that impact the properties of GCL, such as polymer loading, mass per unit area, leachate properties (e.g. monovalent and divalent cations, concentration of Cl and SO 4 , ionic strength, relative abundance of monovalent to divalent cations, anion ratio), effective confining stress, and swell index. The dataset was randomly split, with 80% used to train machine learning models and 20% for evaluating model performance. Seven machine learning algorithms (Linear Regression, Decision Trees, Gene Expression Programming (GEP), Logistic Regression, eXtreme Gradient Boosting (XGBoost), Support Vector Machine, and Artificial Neural Network) were implemented and compared across nine subsets, each representing a combination of different features. The results indicated that XGBoost consistently outperformed other methods in all feature subsets for both regression and classification analyses. The GEP-developed equation was developed to predict the hydraulic conductivity of the B-P GCL to leachates. Classification trees derived from this study can serve as simple tools to screen B-P GCLs for leachate management in solid waste disposal facilities and impoundment ponds.